openvm_rv32im_circuit/mulh/
core.rs

Help
1use std::{
2    array,
3    borrow::{Borrow, BorrowMut},
4};
5
6use openvm_circuit::arch::{
7    AdapterAirContext, AdapterRuntimeContext, MinimalInstruction, Result, VmAdapterInterface,
8    VmCoreAir, VmCoreChip,
9};
10use openvm_circuit_primitives::{
11    bitwise_op_lookup::{BitwiseOperationLookupBus, SharedBitwiseOperationLookupChip},
12    range_tuple::{RangeTupleCheckerBus, SharedRangeTupleCheckerChip},
13};
14use openvm_circuit_primitives_derive::AlignedBorrow;
15use openvm_instructions::{instruction::Instruction, LocalOpcode};
16use openvm_rv32im_transpiler::MulHOpcode;
17use openvm_stark_backend::{
18    interaction::InteractionBuilder,
19    p3_air::{AirBuilder, BaseAir},
20    p3_field::{Field, FieldAlgebra, PrimeField32},
21    rap::BaseAirWithPublicValues,
22};
23use serde::{Deserialize, Serialize};
24use serde_big_array::BigArray;
25use strum::IntoEnumIterator;
26
27#[repr(C)]
28#[derive(AlignedBorrow)]
29pub struct MulHCoreCols<T, const NUM_LIMBS: usize, const LIMB_BITS: usize> {
30    pub a: [T; NUM_LIMBS],
31    pub b: [T; NUM_LIMBS],
32    pub c: [T; NUM_LIMBS],
33
34    pub a_mul: [T; NUM_LIMBS],
35    pub b_ext: T,
36    pub c_ext: T,
37
38    pub opcode_mulh_flag: T,
39    pub opcode_mulhsu_flag: T,
40    pub opcode_mulhu_flag: T,
41}
42
43#[derive(Copy, Clone, Debug)]
44pub struct MulHCoreAir<const NUM_LIMBS: usize, const LIMB_BITS: usize> {
45    pub bitwise_lookup_bus: BitwiseOperationLookupBus,
46    pub range_tuple_bus: RangeTupleCheckerBus<2>,
47}
48
49impl<F: Field, const NUM_LIMBS: usize, const LIMB_BITS: usize> BaseAir<F>
50    for MulHCoreAir<NUM_LIMBS, LIMB_BITS>
51{
52    fn width(&self) -> usize {
53        MulHCoreCols::<F, NUM_LIMBS, LIMB_BITS>::width()
54    }
55}
56impl<F: Field, const NUM_LIMBS: usize, const LIMB_BITS: usize> BaseAirWithPublicValues<F>
57    for MulHCoreAir<NUM_LIMBS, LIMB_BITS>
58{
59}
60
61impl<AB, I, const NUM_LIMBS: usize, const LIMB_BITS: usize> VmCoreAir<AB, I>
62    for MulHCoreAir<NUM_LIMBS, LIMB_BITS>
63where
64    AB: InteractionBuilder,
65    I: VmAdapterInterface<AB::Expr>,
66    I::Reads: From<[[AB::Expr; NUM_LIMBS]; 2]>,
67    I::Writes: From<[[AB::Expr; NUM_LIMBS]; 1]>,
68    I::ProcessedInstruction: From<MinimalInstruction<AB::Expr>>,
69{
70    fn eval(
71        &self,
72        builder: &mut AB,
73        local_core: &[AB::Var],
74        _from_pc: AB::Var,
75    ) -> AdapterAirContext<AB::Expr, I> {
76        let cols: &MulHCoreCols<_, NUM_LIMBS, LIMB_BITS> = local_core.borrow();
77        let flags = [
78            cols.opcode_mulh_flag,
79            cols.opcode_mulhsu_flag,
80            cols.opcode_mulhu_flag,
81        ];
82
83        let is_valid = flags.iter().fold(AB::Expr::ZERO, |acc, &flag| {
84            builder.assert_bool(flag);
85            acc + flag.into()
86        });
87        builder.assert_bool(is_valid.clone());
88
89        let b = &cols.b;
90        let c = &cols.c;
91        let carry_divide = AB::F::from_canonical_u32(1 << LIMB_BITS).inverse();
92
93        // Note b * c = a << LIMB_BITS + a_mul, in order to constrain that a is correct we
94        // need to compute the carries generated by a_mul.
95        let a_mul = &cols.a_mul;
96        let mut carry_mul: [AB::Expr; NUM_LIMBS] = array::from_fn(|_| AB::Expr::ZERO);
97
98        for i in 0..NUM_LIMBS {
99            let expected_limb = if i == 0 {
100                AB::Expr::ZERO
101            } else {
102                carry_mul[i - 1].clone()
103            } + (0..=i).fold(AB::Expr::ZERO, |ac, k| ac + (b[k] * c[i - k]));
104            carry_mul[i] = AB::Expr::from(carry_divide) * (expected_limb - a_mul[i]);
105        }
106
107        for (a_mul, carry_mul) in a_mul.iter().zip(carry_mul.iter()) {
108            self.range_tuple_bus
109                .send(vec![(*a_mul).into(), carry_mul.clone()])
110                .eval(builder, is_valid.clone());
111        }
112
113        // We can now constrain that a is correct using carry_mul[NUM_LIMBS - 1]
114        let a = &cols.a;
115        let mut carry: [AB::Expr; NUM_LIMBS] = array::from_fn(|_| AB::Expr::ZERO);
116
117        for j in 0..NUM_LIMBS {
118            let expected_limb = if j == 0 {
119                carry_mul[NUM_LIMBS - 1].clone()
120            } else {
121                carry[j - 1].clone()
122            } + ((j + 1)..NUM_LIMBS)
123                .fold(AB::Expr::ZERO, |acc, k| acc + (b[k] * c[NUM_LIMBS + j - k]))
124                + (0..(j + 1)).fold(AB::Expr::ZERO, |acc, k| {
125                    acc + (b[k] * cols.c_ext) + (c[k] * cols.b_ext)
126                });
127            carry[j] = AB::Expr::from(carry_divide) * (expected_limb - a[j]);
128        }
129
130        for (a, carry) in a.iter().zip(carry.iter()) {
131            self.range_tuple_bus
132                .send(vec![(*a).into(), carry.clone()])
133                .eval(builder, is_valid.clone());
134        }
135
136        // Check that b_ext and c_ext are correct using bitwise lookup. We check
137        // both b and c when the opcode is MULH, and only b when MULHSU.
138        let sign_mask = AB::F::from_canonical_u32(1 << (LIMB_BITS - 1));
139        let ext_inv = AB::F::from_canonical_u32((1 << LIMB_BITS) - 1).inverse();
140        let b_sign = cols.b_ext * ext_inv;
141        let c_sign = cols.c_ext * ext_inv;
142
143        builder.assert_bool(b_sign.clone());
144        builder.assert_bool(c_sign.clone());
145        builder
146            .when(cols.opcode_mulhu_flag)
147            .assert_zero(b_sign.clone());
148        builder
149            .when(cols.opcode_mulhu_flag + cols.opcode_mulhsu_flag)
150            .assert_zero(c_sign.clone());
151
152        self.bitwise_lookup_bus
153            .send_range(
154                AB::Expr::from_canonical_u32(2) * (b[NUM_LIMBS - 1] - b_sign * sign_mask),
155                (cols.opcode_mulh_flag + AB::Expr::ONE) * (c[NUM_LIMBS - 1] - c_sign * sign_mask),
156            )
157            .eval(builder, cols.opcode_mulh_flag + cols.opcode_mulhsu_flag);
158
159        let expected_opcode = VmCoreAir::<AB, I>::expr_to_global_expr(
160            self,
161            flags.iter().zip(MulHOpcode::iter()).fold(
162                AB::Expr::ZERO,
163                |acc, (flag, local_opcode)| {
164                    acc + (*flag).into() * AB::Expr::from_canonical_u8(local_opcode as u8)
165                },
166            ),
167        );
168
169        AdapterAirContext {
170            to_pc: None,
171            reads: [cols.b.map(Into::into), cols.c.map(Into::into)].into(),
172            writes: [cols.a.map(Into::into)].into(),
173            instruction: MinimalInstruction {
174                is_valid,
175                opcode: expected_opcode,
176            }
177            .into(),
178        }
179    }
180
181    fn start_offset(&self) -> usize {
182        MulHOpcode::CLASS_OFFSET
183    }
184}
185
186pub struct MulHCoreChip<const NUM_LIMBS: usize, const LIMB_BITS: usize> {
187    pub air: MulHCoreAir<NUM_LIMBS, LIMB_BITS>,
188    pub bitwise_lookup_chip: SharedBitwiseOperationLookupChip<LIMB_BITS>,
189    pub range_tuple_chip: SharedRangeTupleCheckerChip<2>,
190}
191
192impl<const NUM_LIMBS: usize, const LIMB_BITS: usize> MulHCoreChip<NUM_LIMBS, LIMB_BITS> {
193    pub fn new(
194        bitwise_lookup_chip: SharedBitwiseOperationLookupChip<LIMB_BITS>,
195        range_tuple_chip: SharedRangeTupleCheckerChip<2>,
196    ) -> Self {
197        // The RangeTupleChecker is used to range check (a[i], carry[i]) pairs where 0 <= i
198        // < 2 * NUM_LIMBS. a[i] must have LIMB_BITS bits and carry[i] is the sum of i + 1
199        // bytes (with LIMB_BITS bits). BitwiseOperationLookup is used to sign check bytes.
200        debug_assert!(
201            range_tuple_chip.sizes()[0] == 1 << LIMB_BITS,
202            "First element of RangeTupleChecker must have size {}",
203            1 << LIMB_BITS
204        );
205        debug_assert!(
206            range_tuple_chip.sizes()[1] >= (1 << LIMB_BITS) * 2 * NUM_LIMBS as u32,
207            "Second element of RangeTupleChecker must have size of at least {}",
208            (1 << LIMB_BITS) * 2 * NUM_LIMBS as u32
209        );
210
211        Self {
212            air: MulHCoreAir {
213                bitwise_lookup_bus: bitwise_lookup_chip.bus(),
214                range_tuple_bus: *range_tuple_chip.bus(),
215            },
216            bitwise_lookup_chip,
217            range_tuple_chip,
218        }
219    }
220}
221
222#[repr(C)]
223#[derive(Clone, Debug, Serialize, Deserialize)]
224pub struct MulHCoreRecord<T, const NUM_LIMBS: usize, const LIMB_BITS: usize> {
225    pub opcode: MulHOpcode,
226    #[serde(with = "BigArray")]
227    pub a: [T; NUM_LIMBS],
228    #[serde(with = "BigArray")]
229    pub b: [T; NUM_LIMBS],
230    #[serde(with = "BigArray")]
231    pub c: [T; NUM_LIMBS],
232    #[serde(with = "BigArray")]
233    pub a_mul: [T; NUM_LIMBS],
234    pub b_ext: T,
235    pub c_ext: T,
236}
237
238impl<F: PrimeField32, I: VmAdapterInterface<F>, const NUM_LIMBS: usize, const LIMB_BITS: usize>
239    VmCoreChip<F, I> for MulHCoreChip<NUM_LIMBS, LIMB_BITS>
240where
241    I::Reads: Into<[[F; NUM_LIMBS]; 2]>,
242    I::Writes: From<[[F; NUM_LIMBS]; 1]>,
243{
244    type Record = MulHCoreRecord<F, NUM_LIMBS, LIMB_BITS>;
245    type Air = MulHCoreAir<NUM_LIMBS, LIMB_BITS>;
246
247    #[allow(clippy::type_complexity)]
248    fn execute_instruction(
249        &self,
250        instruction: &Instruction<F>,
251        _from_pc: u32,
252        reads: I::Reads,
253    ) -> Result<(AdapterRuntimeContext<F, I>, Self::Record)> {
254        let Instruction { opcode, .. } = instruction;
255        let mulh_opcode = MulHOpcode::from_usize(opcode.local_opcode_idx(MulHOpcode::CLASS_OFFSET));
256
257        let data: [[F; NUM_LIMBS]; 2] = reads.into();
258        let b = data[0].map(|x| x.as_canonical_u32());
259        let c = data[1].map(|y| y.as_canonical_u32());
260        let (a, a_mul, carry, b_ext, c_ext) = run_mulh::<NUM_LIMBS, LIMB_BITS>(mulh_opcode, &b, &c);
261
262        for i in 0..NUM_LIMBS {
263            self.range_tuple_chip.add_count(&[a_mul[i], carry[i]]);
264            self.range_tuple_chip
265                .add_count(&[a[i], carry[NUM_LIMBS + i]]);
266        }
267
268        if mulh_opcode != MulHOpcode::MULHU {
269            let b_sign_mask = if b_ext == 0 { 0 } else { 1 << (LIMB_BITS - 1) };
270            let c_sign_mask = if c_ext == 0 { 0 } else { 1 << (LIMB_BITS - 1) };
271            self.bitwise_lookup_chip.request_range(
272                (b[NUM_LIMBS - 1] - b_sign_mask) << 1,
273                (c[NUM_LIMBS - 1] - c_sign_mask) << ((mulh_opcode == MulHOpcode::MULH) as u32),
274            );
275        }
276
277        let output = AdapterRuntimeContext::without_pc([a.map(F::from_canonical_u32)]);
278        let record = MulHCoreRecord {
279            opcode: mulh_opcode,
280            a: a.map(F::from_canonical_u32),
281            b: data[0],
282            c: data[1],
283            a_mul: a_mul.map(F::from_canonical_u32),
284            b_ext: F::from_canonical_u32(b_ext),
285            c_ext: F::from_canonical_u32(c_ext),
286        };
287
288        Ok((output, record))
289    }
290
291    fn get_opcode_name(&self, opcode: usize) -> String {
292        format!(
293            "{:?}",
294            MulHOpcode::from_usize(opcode - MulHOpcode::CLASS_OFFSET)
295        )
296    }
297
298    fn generate_trace_row(&self, row_slice: &mut [F], record: Self::Record) {
299        let row_slice: &mut MulHCoreCols<_, NUM_LIMBS, LIMB_BITS> = row_slice.borrow_mut();
300        row_slice.a = record.a;
301        row_slice.b = record.b;
302        row_slice.c = record.c;
303        row_slice.a_mul = record.a_mul;
304        row_slice.b_ext = record.b_ext;
305        row_slice.c_ext = record.c_ext;
306        row_slice.opcode_mulh_flag = F::from_bool(record.opcode == MulHOpcode::MULH);
307        row_slice.opcode_mulhsu_flag = F::from_bool(record.opcode == MulHOpcode::MULHSU);
308        row_slice.opcode_mulhu_flag = F::from_bool(record.opcode == MulHOpcode::MULHU);
309    }
310
311    fn air(&self) -> &Self::Air {
312        &self.air
313    }
314}
315
316// returns mulh[[s]u], mul, carry, x_ext, y_ext
317pub(super) fn run_mulh<const NUM_LIMBS: usize, const LIMB_BITS: usize>(
318    opcode: MulHOpcode,
319    x: &[u32; NUM_LIMBS],
320    y: &[u32; NUM_LIMBS],
321) -> ([u32; NUM_LIMBS], [u32; NUM_LIMBS], Vec<u32>, u32, u32) {
322    let mut mul = [0; NUM_LIMBS];
323    let mut carry = vec![0; 2 * NUM_LIMBS];
324    for i in 0..NUM_LIMBS {
325        if i > 0 {
326            mul[i] = carry[i - 1];
327        }
328        for j in 0..=i {
329            mul[i] += x[j] * y[i - j];
330        }
331        carry[i] = mul[i] >> LIMB_BITS;
332        mul[i] %= 1 << LIMB_BITS;
333    }
334
335    let x_ext = (x[NUM_LIMBS - 1] >> (LIMB_BITS - 1))
336        * if opcode == MulHOpcode::MULHU {
337            0
338        } else {
339            (1 << LIMB_BITS) - 1
340        };
341    let y_ext = (y[NUM_LIMBS - 1] >> (LIMB_BITS - 1))
342        * if opcode == MulHOpcode::MULH {
343            (1 << LIMB_BITS) - 1
344        } else {
345            0
346        };
347
348    let mut mulh = [0; NUM_LIMBS];
349    let mut x_prefix = 0;
350    let mut y_prefix = 0;
351
352    for i in 0..NUM_LIMBS {
353        x_prefix += x[i];
354        y_prefix += y[i];
355        mulh[i] = carry[NUM_LIMBS + i - 1] + x_prefix * y_ext + y_prefix * x_ext;
356        for j in (i + 1)..NUM_LIMBS {
357            mulh[i] += x[j] * y[NUM_LIMBS + i - j];
358        }
359        carry[NUM_LIMBS + i] = mulh[i] >> LIMB_BITS;
360        mulh[i] %= 1 << LIMB_BITS;
361    }
362
363    (mulh, mul, carry, x_ext, y_ext)
364}
openvm_rv32im_circuit/mulh/core.rs

openvm_rv32im_circuit/mulh/
core.rs
